Visual anomaly detection aims to identify abnormal objects, but training data for anomalies is often scarce. The paper proposes an agentic ensemble that combines multiple pretrained detectors to improve performance in this setting.
According to the abstract, assembling multiple pretrained detectors can help by leveraging their transferable image scores and anomaly-related information. The title indicates that vision-language models are also part of the ensemble, though the provided abstract does not detail the agentic mechanism.
The abstract is truncated, so specific results and evaluation details are not available from the source. The core idea is that ensembling pretrained models can compensate for limited anomaly examples.